Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

claude-jev-warden

Claude Code PreToolUse quality gate where Jev scores drafts (such as SVG illustrations) before they are written and blocks substandard work.

ProjectprototypeMIT
View on GitHub ↗

Overview

Claude Jev Warden is a quality gate and aesthetic supervisor for Claude Code powered by TypeSafe Jev (typesafe/jev-1.13). Through the PreToolUse hook it intercepts Write, Edit and MultiEdit calls before changes touch disk, evaluates the draft with Jev, and blocks substandard work with exit code 2 while injecting diagnostic feedback into the agent's context.

Its main example is art direction for SVG illustrations: Jev reads the SVG source as text and returns a quality score, a verdict and a named deficiency against an approval threshold (80% in the example). A CLI can audit any file directly without launching Claude Code, and the README shows the raw state-and-questions request. Since Jev reads text only, judgments of visual quality are based on the SVG markup, not rendered images.

What the decision model decides

TypeQuestionAnswers
ScoreHow good is this draft (quality score)?—
ChoiceWhat is the verdict and main deficiency?e.g. approve_dynamic_exquisite · ...

How it works

  1. Claude proposes a Write or Edit
  2. Hook sends the draft to Jev
  3. Compare score with the approval threshold
  4. Block with feedback or allow the write

Strengths

  • Enforced gate rather than a prompt suggestion
  • CLI for standalone audits

Caveats

  • Visual judgments based on SVG text, not rendering
  • Threshold and criteria are subjective

Getting started

node bin/jev-audit.js path/to/illustration.svg

At a glance

Type
Project
Status
prototype
Language
JavaScript
Runtime
Claude Code hook + Node CLI
Has code
yes
License
MIT

Use cases and integrations

How curated lists describe it

  • Real-time quality gate and Art Director Warden for Claude Code powered by TypeSafe Jev 1.13 non-autoregressive decision model

    logicrw/awesome-jev-projects

Frequently asked questions

What does it block?

Writes whose Jev quality score falls below the threshold.

Can Jev see the image?

No, it reads the SVG source as text.

Can I use it outside Claude Code?

Yes, via the jev-audit CLI.

Related projects

  • jev-storyboard-lab

    Companion code comparing Google ADK and Microsoft Agent Framework for structured video timelines, with Jev as a shared per-segment QC gate.

    LibraryprototypePython
  • Boson-Video

    Video summarizer that ties every line to its second; Jev checks each summary sentence against what was said, with numbers checked in code.

    ProjectbetaPython
  • jev_project_context

    Evidence-first experiment-memory skill for coding agents, with an optional Jev audit layer that checks provenance and claim support.

    Agent skillbetaPython
  • jev-enforce

    Claude Code plugin that checks every reply and file edit against your CLAUDE.md rules with one Jev yes/no per rule.

    PluginbetaTypeScript
  • jev-gates

    Claude Code plugin with nine Jev-judged gates for rules, scope, intent, completion and claims that escalate but never approve.

    HookbetaTypeScript
  • slopcheck-jev

    Prose linter for AI writing tells: regex handles 18, Jev judges 15 as Nouls in one call, shipped as a Claude Code Stop hook.

    HookbetaPython